Dead Code Eliminator Agent Skill

Dead Code Eliminator

Overview

Systematically identify unused or redundant code in Python codebases to improve maintainability, reduce confusion, and eliminate technical debt.

Workflow

1. Understand the Scope

Define what to analyze:

Questions to ask:

• What directory or files should be analyzed?
• Should test files be included or excluded?
• Are there specific types of dead code to focus on?
• Should external-facing API functions be considered?

Determine analysis scope:

# Check project structure
ls -la

# Count Python files
find . -name "*.py" | wc -l

# Identify test directories
find . -type d -name "*test*"

2. Identify Dead Code

Use multiple detection strategies to find different types of dead code.

Strategy 1: Find Unused Functions

Use the bundled script for AST-based analysis:

# Scan entire project
python scripts/find_unused_functions.py /path/to/project

# Exclude specific directories
python scripts/find_unused_functions.py /path/to/project tests,venv,docs

What it detects:

• Functions defined but never called
• Methods that aren't invoked anywhere
• Async functions without callers

Limitations:

• Won't detect dynamically called functions (via getattr, decorators)
• May flag public API functions that are used externally
• Doesn't detect pytest fixtures or entry points

Strategy 2: Find Unused Imports

Use the bundled script to identify unused imports:

# Scan single file
python scripts/find_unused_imports.py /path/to/file.py

# Scan entire directory
python scripts/find_unused_imports.py /path/to/project

# Exclude directories
python scripts/find_unused_imports.py /path/to/project venv,.venv,tests

What it detects:

• Imports that are never referenced
• Unused from X import Y statements
• Redundant imports

Strategy 3: Use External Tools

Leverage Python ecosystem tools for comprehensive analysis:

vulture - Finds unused code:

# Install
pip install vulture

# Run on project
vulture /path/to/project

# Exclude directories
vulture /path/to/project --exclude venv,tests

# Set minimum confidence (0-100)
vulture /path/to/project --min-confidence 80

autoflake - Focuses on imports and variables:

# Install
pip install autoflake

# Check for unused imports
autoflake --check --imports /path/to/file.py

# Check unused imports and variables
autoflake --check --remove-all-unused-imports --remove-unused-variables /path/to/file.py

# Recursive scan
autoflake --check -r /path/to/project

pylint - General linting including dead code:

# Install
pip install pylint

# Check for unused variables, imports, functions
pylint /path/to/project --disable=all --enable=unused-import,unused-variable,unreachable

Strategy 4: Manual Code Review

Read the code to identify patterns:

Unreachable code:

• Code after return statements
• Code in impossible conditions
• Code after break, continue, or raise

Redundant conditions:

• Always-true or always-false checks
• Duplicate conditions
• Unnecessary else after return

Look for:

# Find code after return statements (basic pattern)
grep -A 3 "return" **/*.py | grep -v "^--$"

# Find functions with "old" or "legacy" in name
grep -r "def.*old\|def.*legacy" .

# Find TODO comments about removal
grep -r "TODO.*remove\|FIXME.*delete" .

3. Categorize Findings

Organize dead code by type and priority.

See dead-code-patterns.md for comprehensive pattern catalog.

Category: Unused Imports

Priority: High (easy to remove, low risk)

Examples:

• import os but os is never used
• from typing import List, Dict but only List is used
• Duplicate imports

Category: Unused Functions/Methods

Priority: Medium to High

Subcategories:

• Orphaned helpers: Utility functions never called
• Refactoring leftovers: Old implementations not removed
• Test helpers: Test utilities not used by any test

Caution - May be intentional:

• Public API functions (used externally)
• Plugin/hook functions (called dynamically)
• CLI entry points (called from command line)

Category: Unreachable Code

Priority: High (indicates bugs or confusion)

Examples:

• Code after return
• Code in impossible conditions
• Code after raise

Category: Redundant Code

Priority: Medium

Examples:

• Redundant boolean checks
• Unnecessary else after return
• Duplicate logic in multiple places

Category: Unused Variables

Priority: Low to Medium

Examples:

• Assigned but never read
• Function parameters never used
• Loop variables never referenced

4. Verify Findings

Before reporting, verify that identified code is truly dead.

Check for dynamic usage:

# Code may appear unused but is called dynamically
handlers = {
    'process': process_handler,  # Looks unused but isn't
}

# Or via getattr
handler = getattr(module, function_name)

Check for external usage:

• Is this a public API function?
• Is it documented in README or API docs?
• Is it an entry point in setup.py?

Check for framework conventions:

# Django signal handlers
@receiver(post_save, sender=User)
def user_saved(sender, instance, **kwargs):  # May appear unused
    pass

# Pytest fixtures
@pytest.fixture
def sample_data():  # Used by tests but not "called" directly
    return {"key": "value"}

Verify with grep:

# Search for function name in entire codebase
grep -r "function_name" .

# Search in quotes (dynamic calls)
grep -r "'function_name'\|\"function_name\"" .

# Search in setup.py or config files
grep -r "function_name" setup.py pyproject.toml

5. Generate Report

Create a structured markdown report of findings.

Dead Code Analysis Report

Project: [Project Name]

Analyzed: [Date]

Scope: [Directories analyzed]

Excluded: [Excluded directories]

Summary

• Unused imports: X findings across Y files
• Unused functions: A findings across B files
• Unreachable code: M findings
• Redundant code: N findings

Total: Z dead code instances found

🔴 High Priority Issues

Unreachable Code

Code that can never execute - should be removed immediately.

Issue 1: Code After Return

Location: src/utils.py:45-47

Code:

def process_data(data):
    if not data:
        return None
        logging.warning("Empty data")  # UNREACHABLE
        validate(data)  # UNREACHABLE

Recommendation: Remove lines 46-47 (unreachable after return).

Impact: Misleading code that suggests validation happens but doesn't.

Unused Imports

File: src/main.py

Lines:

• Line 3: import os (unused)
• Line 5: from typing import Dict, Tuple (only Dict is used)
• Line 12: import re (unused)

Recommendation: Remove unused imports. Update line 5 to from typing import Dict.

🟡 Medium Priority Issues

Unused Functions

Functions that appear unused but should be verified before removal.

Issue 3: Orphaned Helper Function

Location: src/helpers.py:89

Function: format_timestamp(ts: int) -> str

Analysis:

• Defined but never called in codebase
• Not in public API documentation
• Not an entry point

Verification needed:

• ✅ Checked: Not in setup.py entry_points
• ✅ Checked: Not mentioned in README
• ✅ Checked: No string references in codebase
• ❌ Need to verify: Could this be used by external code?

Recommendation: If not part of public API, remove. Otherwise, document it.

Issue 4: Duplicate Logic

Locations:

• src/processor_a.py:45-52
• src/processor_b.py:78-85

Code: Both files contain identical validation logic.

Recommendation: Extract common logic into shared utility function.

Impact: Maintenance burden - changes must be duplicated.

🔵 Low Priority Issues

Redundant Code

Issue 5: Unnecessary Else After Return

Location: src/validator.py:123-127

Code:

def check_status(value):
    if value > 0:
        return "positive"
    else:  # Unnecessary
        return "negative"

Recommendation: Remove else clause (implicit after return).

Impact: Minor - slightly less readable but no functional impact.

Issue 6: Unused Variable

Location: src/calculator.py:56

Code:

def compute(a, b):
    total = a + b  # Assigned but never used
    return a * b

Recommendation: Remove unused total variable.

⚪ Needs Verification

Potentially Used Dynamically

These appear unused but may be called dynamically. Manual verification needed.

Function: handle_create()

Location: src/handlers.py:34

Reason for caution: File contains handler registry suggesting dynamic dispatch.

Code pattern:

HANDLERS = {
    'create': handle_create,
    'update': handle_update,
}

Recommendation: Verify this is registered and used. If confirmed unused, remove.

Recommendations

Immediate Actions (High Priority)

• Remove unreachable code (Issue 1)
• Clean up unused imports across all files
• Verify and remove orphaned helper functions

Short-term Actions (Medium Priority)

• Extract duplicate logic into shared utilities
• Verify dynamically-called functions
• Remove confirmed unused functions

Long-term Actions (Low Priority)

• Simplify redundant conditions
• Remove unused variables
• Establish linting rules to prevent future dead code

Prevention

Add to CI/CD pipeline:

# Add to pre-commit hook or CI
vulture src/ --min-confidence 80
autoflake --check -r src/

Configure IDE:

• Enable unused import warnings
• Configure pylint/flake8 for dead code detection

Code review checklist:

• [ ] No unused imports?
• [ ] No unreachable code?
• [ ] No unused variables?
• [ ] Functions have callers?

Detailed Findings

[If needed, include full lists of all findings organized by file]

6. Present Findings

Share the report with the team and get feedback.

Present clearly:

• Start with summary statistics
• Highlight high-priority issues first
• Provide specific locations and recommendations
• Separate definite dead code from "needs verification"

Be cautious about:

• Public API functions that may be used externally
• Dynamic dispatch patterns
• Framework-specific code (decorators, fixtures, signals)
• CLI entry points
• Plugin systems

Request feedback:

• "Are these functions part of the public API?"
• "Is this code used by external tools or scripts?"
• "Should we keep this for planned features?"

Tips for Effective Dead Code Analysis

Start conservatively:

• Focus on obvious cases first (unused imports, unreachable code)
• Be cautious with unused functions (may be called externally)
• Verify before removing

Use multiple detection methods:

• AST-based analysis (bundled scripts)
• External tools (vulture, autoflake, pylint)
• Manual code review
• Coverage analysis (find untested code)

Prioritize by risk:

• Low risk: Unused imports, unused variables
• Medium risk: Unused internal functions, redundant code
• Higher risk: Functions that might be public API

Consider the context:

• Age of code (old = more likely truly dead)
• Recent refactoring (may be leftovers)
• Project type (library vs application)
• Team size (more people = more likely to have external usage)

Prevention is better than cure:

• Enable linting in CI/CD
• Configure IDE warnings
• Code review checklist
• Regular dead code audits

Common False Positives

Be aware of code that appears dead but isn't:

1. Dynamic dispatch:

handler = getattr(module, f"handle_{action}")

2. Entry points:

# setup.py
entry_points={
    'console_scripts': ['tool=module:main_function']
}

3. Pytest fixtures:

@pytest.fixture
def sample_data():  # Used by tests implicitly
    return data

4. Django signals:

@receiver(post_save, sender=Model)
def handle_save(sender, instance, **kwargs):  # Called by framework
    pass

5. Decorators and metaclasses:

class Meta:
    def __init_subclass__(cls):  # Called implicitly
        register(cls)

6. Public API functions:

# In library code - may be used by external code
def public_function():  # Appears unused internally
    pass

Reference

For comprehensive dead code patterns and detection strategies, see dead-code-patterns.md.